Fix "wrong" read method called

This is conceptually a bit wrong as it is since I guess it would break
on GZipped responses, though they would not occur here due to the Git
protocol.

In the urllib3 case, however, it is more pressing since the
resp.read method is always empty at this stage due to the data already
having been consumed in _http_request (unless preload_content=False
is given).

Use the read method prepared and provided by _http_request instead.